Help explain to me this L2 Cache, please.

[quicksilver17]

I will be purchasing a new pc, with the specs of this:

ASPIRE ATX-AS600W-BL ATX12V/ EPS12V 600W Power Supply,

ASUS A8N-SLI Premium Socket 939 NVIDIA nForce4 SLI ATX AMD Motherboard

eVGA Geforce 7800GTX

Seagate Barracuda 7200.8 ST3250823AS 250GB 7200 RPM Serial ATA150 Hard Drive

Thermaltake Tenor VB2000BNS Black Aluminum/Steel ATX Desktop Computer Case

now for my processor I have narrow it down to two:

AMD Athlon 64 X2 4600+ Manchester
or the
AMD Athlon 64 X2 4800+ Toledo

With the only difference is the L2 Cache. Will I really need an extra 1md of l2 cache if I go with the 4800+? I will be both gaming and multiprocessing quit a bit(Well not at the same time,though). Also could you please define more of the L2 Cache. I have the basic idea of, just cant really grasp it.

Thx for those who can help.

P.S.> If anyone at the top of their head could think of some good memory ram to go with this beast, that would be helpful too.

I almost forgot, read somewhere on this forum, that more cache help OCing easier, How so? I will be OCing, but with the AI doing it for me from Asus. Maybe later on in the future, dont want to burn out the cpu/memory.<P ID="edit"><FONT SIZE=-1><EM>Edited by quicksilver17 on 09/07/05 03:43 PM.</EM></FONT></P>

 

[tendercare]

I have a follow up question: with all other variables the same, is performance proportionate to L2 size?

 

[marneus]

Generally the more the better,
cache sits between the processor & the system memory
level cache 1 is the fastest, level 2 is slower than level 1 but still faster than system memory, intel added level 3 cache in the order of several megabytes, up to 18mb in some cases...

Trust me I know what I'm doing... ooops, grab the cat...

 

[ChipDeath]

To answer your question in the most basic terms:
You almost certainly wouldn't notice a difference between the two.

AMD's architecture doesn't benifit nearly as much as intel's from increasing cache sizes anyway. The <i>only</i> situation where you'd see a very large gain would be in running very small operations which don't <i>quite</i> fit entirely inside the L2 cache of the smaller version, but do in the larger version (as then it avoids the 'waiting' while it fetches stuff from memory after it's been 'around the loop' once). Since these sorts of things are usually only tiny benchmarking programs it's not something to worry about really.

Certainly don't get the 4800+ if that means downgrading some other component (RAM, gfx card, etc). If you have the cash spare though, why not? .5% extra performance is still extra performance...

As for overclocking, I'm not sure. I've seen people think that the larger surface area of the die when it has more cache helps with cooling (therefore overclocking), but then there's also more transistors so I wouldn't have thought it would make that much difference (if any). It's pretty much pot luck as to how any particular core will OC. I haven't read up much on it, but I've seen others say the 1Mb cache variants OC better, so maybe they're just all good examples.

As for RAM, you could do a lot worse than some Crucial Ballistix.

---
<pre> (\_/)
|~~~~~|======
|_____| This was bunny. He was tasty.
/\/\/\/\</pre><p>

 

[slvr_phoenix]

While ChipDeath's explanation wasn't bad, I'll add my own anway, mostly because I just uploaded to my source code repository a file I've been working on and I could use a break now.

Anybunny, cache is like your RAM, but <i>way</i> faster, and a lot smaller. What cache does is let the CPU hold a small part of the memory in a place that's quick to access, so that it doesn't always have to take forever by accessing your RAM for every little thing. Why? Because compared to the speeds of the CPU, RAM is very slow.

The L1 cache is the fastest, but also the smallest because fast has its limitations. The L2 is slower than the L1, but still way faster than your RAM. (And some expensive CPUs will have a further L3 cache, same concept. Bigger but slower than previous L, but still faster than RAM.)

When your proc needs something from RAM it first asks the L1 if there's a copy of it there. If there is, time saved. If not, it then asks the L2. If the L2 has it, again, time saved. If not, it then proceeds to go through the long and arduous process of asking the RAM for the data it needs.

So the larger your caches are (L2 is the most common to change the size of) the less often your CPU will need to sit around bored while it goes all of the way to your RAM for the data it needs. ...In theory.

In reality, these things always have points of deminishing returns. Sometimes things aren't stored all nice and ordered in memory, so they're hard to cache this way and thus you have to hit the RAM a lot anyway. Other times processors can pre-determine and pre-fetch what memory they'll need accurately, allowing them to grab it before they need it, thus saving them from the normal lags of RAM. And often everything that a program needs fits into a nice little package so a larger L2 just won't make a difference.

So while in theory more L2 is better, in practice you often don't see the kinds of performance gains from L2 increases that AMD credits the part numbers with. This is <i>especially</i> true on Athlon64s and such because their on-die memory controller speeds up the processor's access of your RAM considerably, thus reducing the speed difference between accessing the L2 cache and accessing the RAM.

So in your case, is the larger L2 cache really worth the extra money? Not really. Some applications might do better for it, but in general you won't notice the difference.

یί∫υєг ρђœŋίχ
<font color=red><i>Deal with the Devil. He buys in bulk.</i></font color=red>
@ 197K of 200K!